WhatWeRun, a database of real production AI agent stacks (whatwerun.dev)

WhatWeRun, a newly launched public database, catalogues real AI agent stacks currently in production across various organizations. This resource compiles insights from engineering blogs, HN threads, and personal narratives, maintaining the contextual accuracy of costs and operational details. Highlighted stacks include Bkper’s AI gateway for bookkeeping, which combines narrow AI capabilities for document classification, and Pinterest's tuneable Analytics Agent, which automates data inquiries and improves efficiency via a feedback loop. The database aims to provide a comprehensive view of how established companies are leveraging AI agents to optimize workflows and automate technical processes. The significance of WhatWeRun lies in its transparency and accessibility, offering the AI/ML community an invaluable reference for real-world implementations. By sharing diverse case studies—from a security-hardened Kubernetes runtime developed by NVIDIA to autonomous coding agents at Stripe—the database facilitates knowledge sharing and innovation. This repository not only showcases the current landscape of AI agent technology but also serves as a springboard for developers and organizations seeking to enhance their own AI capabilities by learning from others’ successes and challenges.